高超声速飞行器热防护系统设计中高精度气动热分析模型使得设计计算成本不断增加，基于数据驱动的气动热环境预示方法受到广泛关注。针对有限高精度模型计算成本下提升全局预示精度的问题，提出一种基于模糊聚类的批量自适应采样方法。根据预示误差分布特征通过聚类采用超球分割构建采样影响域，兼顾误差较大的重点采样域与全局探索；通过当地误差评分系数加权构建采样拒绝域，减小新增样本冗余；结合maxmin准则在综合确定的重点采样空间中新增样本，提升采样质量，进而实现预示模型全局精度快速提升。数值测试算例表明，所提方法与One-Shot、APSFC、CV-Voronoi方法相比能有效降低所需采样规模，加速提升预示精度。通过类HTV-2飞行器气动热快速预示实例，验证了方法的有效性与工程实用性。

High-fidelity aerothermodynamics analysis models in the thermal protection system of the hypersonic vehicles significantly increase the computational budget of engineering design， drawing extensive attention on data-driven based rapid prediction methods. This paper proposes a batch adaptive sampling method based on fuzzy clustering to improve the global prediction accuracy with the limited computational budget of high fidelity models. The sampling influence domain is constructed by clustering and hypersphere segmentation with the distribution characteristics of the prediction error， considering both the key sampling domain with larger errors and global exploration. The sampling refused domain is developed by the local error scoring coefficient weighted to reduce the redundancy of newly added samples. The method adds new samples in the comprehensively determined key sampling space to improve the sampling quality based on the maxmin criterion， thereby rapidly improving the global accuracy of the prediction models. The comparison results show that the proposed method outperforms One-Shot， APSFC and CV-Voronoi in terms of reducing the sampling scale required and accelerating prediction accuracy improvement. The rapid prediction results of the HTV-2 typed vehicle aerothermodynamics demonstrate the practicality and effectiveness of the proposed method in engineering practices.